Part 2: The Formula for Success
In Part 1: Focus on the Solution, I overviewed the situation enterprises are challenged with when implementing AI, and specifically NLP, successfully. The age-old question of build vs. buy is the first decision teams need to make as they consider solving a need. I finished by proposing that the solution may not be exclusively build vs. buy, but a way that empowers businesses to ‘build’ on a platform to deliver results.
Continuing with the goal of building a successful solution that scales and stands the test of time, you don’t have to choose between build vs. buy—you can do both, all while yielding better results. Here are three main reasons why:
1. Time to Value
Most organizations looking to use AI, specifically NLP and NLU, are doing so with the objective of increasing efficiency, lowering costs, and mitigating risk. They have the mission of improving the bottom line and empowering their employees to be more efficient. This is mission critical, and a task not to be taken lightly. Given the buzz around Large Language Models (LLMs) put out by Facebook, OpenAI, Google and other sources, an NLP/NLU project may sound simple. You may think, “hey, why don’t we just use these models.” In reality, a tremendous amount of work goes into the development process and into putting these models in production.
Here is just a sample of what it takes to create your own NLP models:
Identify the data > prepare the data > find the appropriate algorithm > create a workflow that ingests the data into the algorithm > annotate the data > tune the model > compare models for performance > deploy the model into a production workflow (with data ingestion from various sources, pre- and post- processing, and then determine what to do with the results, where it should go, how to view it, who sees it and what they do with it) > manage and monitor the model > create a feedback loop for continuous improvement.
If you plan to use an LLM, once you get results, you’ll still need to answer the question: are the results good enough?
Out of all these steps, there are dozens of tools, libraries and services that will help perform these tasks, and do it well. However, someone still must put it all together. Remember that 85% of AI projects don’t make it into production? It’s because orchestrating all of these tasks individually is incredibly complex; this is what leads to the 85% failure rate.
By deploying a purpose-built NLP platform—not a compilation of open-source tools, but a platform built from the ground up to allow users to build, optimize and deploy NLP models into production—a business can ensure that the foundation of capabilities, the integration of different functions, and the usability is rock solid and ready to use.
When an organization is on a mission to solve a problem and create value for the business, our experience shows that their time to value can be exponentially decreased by deploying a platform that is ready for teams to deliver value immediately by skipping the plumbing headaches, the data integration issues and the data preparation tasks.
This brings us to our second point, around resources.
2. Maximize resource value
Organizations that work with expert.ai usually have multiple groups working together on projects. These groups include business owners, data scientists, and technology teams. Everybody, rightfully so, wants to participate in the project and add value, and they should!
Our clients that have the shortest time to their MLP (minimum lovable product), and the greatest success allocate their resources for maximum value. They’re able to do this once they deploy a platform that empowers them to build and deploy the NLP models into production as quickly and effectively as possible.
So, where should each team focus?
The business teams should focus on driving the KPIs so they can prove that the business problem is being solved. They should be empowered to monitor the KPIs and have a feedback loop to the Data Science and Technology teams to voice their needs, their areas for improvement, as well as the wins they see. If we want to ensure that the business stays in the world of the “15% of projects [that] succeed,” they need to see results fast and communicate with the other teams for continuous improvement. If there is no communication, or requests back to the Data Science and Technology teams fall on deaf ears because they can’t effectively or efficiently adapt their technology to the changing business requirements, this is a reason that the progress vector veers off course. We’ve seen it happen.
The Data Science teams should focus on model training, model creation, and model optimization, in the context of ensuring the results are focused on solving business problems. These three aspects of the workflow are incredibly important for any machine learning project, and NLP/NLU is no different. Specifically, these teams need to focus on:
- Data Annotation
- What to annotate?
- How it should be annotated?
- Model Creation and Performance Analysis
- What algorithms should we use?
- What parameters and hyper-parameters should we use?
- Which models yield the best results for Precision, Recall, and F-Measure?
- Are we accomplishing the business objectives?
- Model Optimization
- How is the model doing in production?
- How is the model doing in production over time?
- What feedback are we receiving that we can use to enhance the model?
- What are the changing business requirements, and how do we ensure the models stay aligned to the business goals?
Our experience shows that when Data Scientists are provided with a platform that empowers them to focus on these high-value questions and allows them to deliver results without worrying about other factors, such as whether a specific function is going to work, or whether an integration is needed for a specific algorithm and so on…businesses see the best results. In an enterprise environment where competition is ripe with innovation, speed and efficiency, results are what will propel you forward.
The third team that is involved is the Technology and IT (Information Technology) group. Build vs. buy becomes a consideration with these teams most frequently because, with all the open-source tools out in the wild, why wouldn’t a team of knowledgeable, smart, and capable individuals want to build a solution? And again, referring to an earlier point, the discussion is NOT around whether they CAN build it, it’s around whether they SHOULD. When a business sets out to solve a business problem, or to increase efficiencies, or to mitigate risk, we need to ensure that the business objective gets solved.
We see organizations trying to solve NLP problems on their own, and what ends up happening is that these highly capable individuals are spending valuable cycles (and payroll dollars) trying to integrate many different components together, trying to research what else they need, and working to debug issues on their own… which could take months if not longer, or if ever achieved at all.
Successful organizations, the ones that land in the 15%, work to maximize their talent and start a project from the 50-yard line (the middle) instead of at the 99-yard line (or at the beginning). Why not provide a platform that gives everybody a valuable head start? Why not let the technology groups focus on finding new ways to integrate NLP into business workflows, building new connectors to more systems that can utilize the models, and deliver even more value to other lines of business?
The goal is to maximize value from talent, ensure that individuals are focused on the highest value tasks, and provide an environment where everybody is working to their fullest potential to maximize business results.
The third, and final, reason why businesses don’t need to choose between build vs. buy…
3. Support, Support, Support
If you’re deploying a CRM that is responsible for predicting revenue to investors, and enables thousands of sales reps to manage their business, it’s crucial that the CRM is working. What if it didn’t work? What if on the last day of the quarter, everybody runs into a ‘404 Not Found’ error when trying to access the CRM? Is it better to have a support number with an established business and a dedicated customer success team that can get you operational as soon as possible versus a few phone numbers for software developers that created the CRM? What if those software developers left the company? The same conversation should be discussed when deploying NLP solutions. This is now a business-critical component, and no longer just a fun-to-have or a science experiment.
When organizations build on their own, they forfeit the ability to call a centralized support number when there are issues. By deploying a purpose-built platform that empowers users to create, develop, test and deploy NLP solutions (such as expert.ai), you know three things:
- It is pre-tested. It will work. It has gone through the proper Quality Assurance (QA) cycles to ensure that it is enterprise ready.
- If you run into issues, there is a whole team dedicated to assisting you 24/7.
- Expert.ai’s sole focus is building a platform for NLP. That’s all we care about, so our customers know it will be future-proof for enhancements, updates, and upgrades.
Circling back to the beginning of this series, expert.ai’s goal is to enable companies to successfully deploy NLP projects into production to help them make money, save money, improve work and mitigate risk.
My personal mission working for expert.ai is to help guide our clients in understanding how to do so in the best possible way. We aren’t new to this; we’ve seen it before. If an organization wants to successfully use AI and NLP to accomplish business goals, then let’s change the conversation from build vs. buy and focus on how we can leverage a proven platform to build up the capabilities you need to drive value and ensure that results will follow.